Which comes first? Unpacking the relationship between peace agreements and peacekeeping missions

Abstract

Why do combatants engaged in civil conflict sign peace agreements when they do? Does a commitment by the United Nations (UN) to send a peacekeeping mission increase the probability that combatants will sign an agreement? With regards to the relationship between peace agreements and UN peacekeeping missions, previous studies of civil war have taken one of two positions: (a) a peace agreement between combatants causes an increase in the probability that the UN will send a peacekeeping mission to the conflict; or (b) the UN’s decision to send a peacekeeping mission to a conflict causes an increase in the probability that combatants will sign a peace agreement. I contend that the UN’s decision to send peacekeepers to a conflict and the combatant’s decision to sign a peace agreement occur simultaneously. To overcome the simultaneity, I exploit exogenous variation in the UN’s willingness to send peacekeepers to conflicts in the mid-1990s, based on the experience of the UNOSOM II mission in Somalia, to identity the causal effects that UN peacekeeping has on the likelihood of combatants signing peace agreements. Not accounting for endogeneity, the data suggest that the UN’s willingness to send peacekeepers increases the probability of a peace agreement. A recursive bivariate probit model using Black Hawk Down as an instrumental variable, however, indicates that the UN’s willingness to send peacekeepers to a conflict does not increase the probability that combatants will sign an agreement. This finding suggests that combatants sign agreements for reasons related internally to the conflict, not pressure from the international community.

Keywords

Civil war conflict termination peacekeeping peace agreements

Introduction

Under what conditions will combatants in civil wars sign peace agreements? What effect does the international community, through the deployment of peacekeepers, have on the probability that combatants will sign an agreement? Understanding the process by which peacekeeping missions are assigned to conflicts is vital to determine the effect of peacekeeping missions on the duration of peace that follows civil wars. If the deployment of peacekeepers is related to the probability of conflict recurrence, then estimates of the effectiveness of peacekeeping operations will be biased. In particular, if the United Nations (UN) deploys peacekeeping missions to conflicts that are not likely to recur, then a comparison of the cases with and without peacekeeping missions will overestimate the impact of peacekeeping missions on peace. On the other hand, if peacekeeping missions go to conflicts that are more likely to recur, then studies will underestimate the impact of peacekeeping missions on peace. Therefore, understanding the factors that make peacekeeping assignment non-random is an important first step in evaluating the success of peacekeeping operations.¹

Furthermore, if there is a positive relationship between peacekeeping missions and peace agreements, it may be impossible to study the effects of peacekeeping missions on post-war peace periods. Most studies of peacekeeping effectiveness use a “post-war period” as their unit of analysis. This period is usually defined as beginning after one side is defeated (if the war ends in military victory) or after the signing of a peace agreement. Fortna (2008: 20) poses the question as “among civil wars that reach some sort of break in the fighting, what distinguishes those that see peacekeeping missions form those that do not?” The problem is that, if the UN’s decision to send peacekeepers to a conflict is what causes the signing of a peace agreement, then some “post-war periods” enter the dataset that otherwise would not have had the UN not been involved. If combatants are signing peace agreements they would not sign otherwise because of UN peacekeepers, then studies of post-war peace may be impossible because no counterfactual “post-war period” without peacekeepers enters the dataset. That is, had the UN not offered its services, the combatants would have continued to fight and the observation would not have entered the dataset as a “post-war period”. The primary motivation of this study is to determine whether or not this is a problem for studies of post-war peace, by examining whether or not the UN’s agreement to send peacekeepers to a conflict affects the probability that an agreement will be signed.

This analysis utilizes an instrumental variable that captures the time immediately after the “Black Hawk Down” incident, which affected the UN’s willingness to send peacekeepers to conflicts, to identify the causal effect of UN peacekeeping on the probability that combatants sign a peace agreement. Fortunately, after accounting for the endogenous nature of peacekeeping deployment, I find that the UN’s willingness to send peacekeepers to a conflict does not increase the probability that combatants will sign a peace agreement.

In order to account for the endogenous nature of peacekeeping missions, one needs an instrumental variable that affects the UN’s willingness to send peacekeepers to a conflict, but does not have any effect on the combatant’s willingness to end the conflict (Angrist and Krueger, 2001). I argue that the Black Hawk Down incident, a highly publicized and embarrassing defeat of US forces by Somali fighters on 3 and 4 October 1993, fits this condition. In an article entitled “UN efforts everywhere turn to dust—Downed helicopter in Somalia doomed a ‘new world order’”, Tom Ashbrook argues that “when the fighting finally ended, an angry mob of Somalis dragged the half-naked body of a US soldier through the streets in front of cameras. As that shocking image was transmitted around the world, the ‘Somalia syndrome’ was born.” Because of this, “conventional wisdom is now that large-scale, UN-led military intervention in the name of peace is unlikely to be seen again soon” (Ashbrook, 1995). For the model to be identified using Black Hawk Down as a shock, it must be true that both: (a) the UN engaged in fewer peacekeeping missions after the incident; and (b) the event had no independent effect on the combatants’ willingness to sign peace agreements, other than through its effect on the UN. I argue below that this exclusion restriction is valid.

Past work on civil war resolution and peacekeeping

Previous studies examining the relationship between peacekeeping and peace agreements fall into two categories. The first category assumes that peacekeepers go where peace agreements have been signed. The second category assumes that peace agreements are signed because a peacekeeping mission is either in the country or has agreed to deploy to a country.

The reason for the contradictory positions taken by scholars probably stems from the fact that peacekeeping missions are deployed in four distinct scenarios: (a) prior to violence (e.g. the UNPREDEP mission in Macedonia); (b) so-called “peacemaking” missions with Chapter VII mandates that deploy during violence (e.g. ONUB in Burundi, which deployed while the government was still fighting the FNL); (c) after a cease-fire between the two sides (e.g. the UNMOT mission in Tajikistan); or (4) after a comprehensive peace agreement (e.g. UNAMIR in Rwanda). The causes of UN intervention in each of these periods of conflict are likely to differ. In particular, deployments that accompany a peace agreement, in which combatants have agreed to a process that will bring about an end to the conflict, are likely to be deployed for fundamentally different reasons than deployments to conflicts without an accompanying peace agreement. Gilligan and Sergenti (2008: 94) make the point that “the data generating process of ongoing wars is quite likely different from that for wars that have ended”. Because of this, the authors separate their analysis of the effectiveness of peacekeeping missions into “in-war” interventions and “post-war” interventions, owing to the profound difference between these two types of interventions.

Do peacekeepers go where peace treaties have been signed?

The first systematic study of the determinants of UN deployment, Gilligan and Stedman (2003), utilizes a Weibull duration model in order to measure the factors that determine when the UN will intervene in a conflict. In their analysis, Gilligan and Stedman include a dummy variable indicating whether the combatants signed a peace treaty to “control for the possibility that the UN is more likely to intervene in a conflict when the combatants have signed a peace treaty” (p. 13). They find no evidence “that the UN is more likely to intervene in a conflict when combatants have negotiated a peace treaty, but this is probably due to multicollinearity” (p. 49). The authors believe that the lack of a significant finding is due to both the correlation between treaty and size of the government’s army variables, and a small sample size.

Fortna (2004) similarly attempts to explain the deployment of international peacekeepers by examining whether or not combatants had signed a peace treaty. She finds that “if peacekeepers deployed where there was ‘peace to keep’ or where the combatants had signaled their ‘political will’ for peace by signing a treaty, we would expect this variable to have a positive effect, especially on consent-based missions … but we can confidently reject the hypothesis that peacekeepers are more likely to intervene when a formal treaty has been signed” (p. 279).²

Mullenbach (2005) approaches the question of where peacekeepers are sent from a slightly different perspective. Instead of focusing solely on the attributes of conflicts such as duration or intensity, Mullenbach wishes to answer the question “what effect do international-level factors have on the likelihood that third-party peacekeeping personnel will be deployed in an intrastate dispute?” (p. 529). Mullenbach codes: (a) whether the country in conflict has a military alliance with a major power; (b) whether the country is itself a major power; (c) whether a major power has been previously involved in the conflict; and (d) whether an international institution has attempted any conflict management activities, such as military sanctions, human rights monitoring or fact-finding missions.

In the analysis, Mullenbach includes whether a ceasefire agreement was signed as a control variable in order to “control for the possibility that the likelihood of a peacekeeping mission varies depending on whether or not the parties formally agree to a cessation of military hostilities” (p. 543). The analysis shows that “the coefficients for ceasefire agreement were positive and statistically significant in each of the eight models … not surprisingly, third-party peacekeeping missions are significantly more likely to be established when the parties to the conflict have formally agreed to a cessation of military hostilities” (p. 548).

Two important points emerge from these three articles. The first is that they all suggest that the deployment of peacekeepers is influenced by whether or not the combatants have signed either ceasefires or peace treaties. The second, and more important, point is that the authors treat the ceasefire or treaty variables as exogenous. That is, by using ceasefire or treaty dummy variables as explanatory variables, the authors are assuming that the combatants’ decision to sign a peace treaty or ceasefire is independent of any possible decision by the UN to agree to enforce an agreement.

Are treaties more likely to be signed when peacekeepers are present?

Other scholars have attempted to answer the question: when and why do combatants sign peace agreements? These scholars assume that combatants are aware of the actions or views of third-party enforcers prior to the signing of peace agreements, and that combatants then decide whether or not to sign a peace treaty based upon this knowledge. Furthermore, the authors assume that the sending of peacekeepers is exogenous.

Walter (2002) divides conflicts into four categories, based on how far along each conflict advanced in the peacemaking process: (a) no negotiation; (b) negotiation; (c) signed bargain; and (d) successful implementation. An ordered logit model then estimates the extent to which the explanatory variables help get the combatants from one stage to the next. Of particular interest here is getting from the negotiation stage to the signed bargain stage.

Walter finds that “once combatants initiated negotiations … only third-party security guarantees, territorial and political pacts, mediation, and nonterritorial goals have a sizable and significant effect on the decision to sign a peace agreement” (p. 78). Furthermore, “third-party security guarantees have the greatest impact on the willingness of combatants to sign peace settlements” (p. 79). Thus, “if a third party offers to verify or enforce demobilization, as the British did in Zimbabwe in 1979 and the United Nations did in Bosnia in 1994, combatants are significantly more likely to sign a peace treaty than if no such offers are made” (p. 80).

Greig and Diehl (2005) examine “the potential impact that a peacekeeping force might have in facilitating a peace agreement between protagonists” (p. 621). The authors gather data on all attempts to mediate conflicts and initiate negotiations, and also on UN and other third-party interventions in civil wars. They find that “peacekeeping had no general impact on mediation or negotiation successes” (p. 640).

DeRouen and Sobek (2004) examine the factors that determine how civil wars end (i.e. with military victory, treaty or truce). The authors run a multinomial logit model that explains whether a conflict ends in government victory, rebel victory, truce or treaty. They include a dummy variable indicating whether the UN was involved in the resolution of the conflict. They conclude that “one of the strongest predictors of the outcome of civil wars is the intervention of the UN. When the UN intervenes in a civil war, it increases the probability of both truce and treaty” (p. 311).

The authors discussed above all treat the presence or promise of peacekeepers as exogenous. Clearly, the logic of this sequence is the exact opposite of the one discussed in the previous section. Table 1 presents a summary of the structure and findings of the six studies discussed above.

Table 1.

The structure and findings of previous research on the relationship between peace agreements and peacekeeping missions

Study	Dependent variable	Independent variable	Finding
Gilligan and Stedman (2003)	Peacekeeping	Peace agreement	No relationship—owing to collinearity with size of government army
Fortna (2004)	Peacekeeping	Peace agreement	No relationship—a better specification of the Fortna model shows a positive relationship
Mullenbach(2005)	Peacekeeping	Ceasefire	Positive relationship
Walter (2002)	Peace agreement	Peacekeeping	Positive relationship
Greig and Diehl (2005)	Peace agreement	Peacekeeping	No relationship
DeRouen and Sobek (2004)	Peace agreement	Peacekeeping	Positive relationship

How are peace agreements signed in practice?

When do we see peace agreements signed, and what is their relationship to an accompanying peacekeeping mission? Following Gilligan and Sergenti (2008) and Fortna (2008) I argue that peacekeeping missions fall into two general categories: “in-war” missions that occur while the government-rebel group dyad is still fighting; and “post-war” mission that occur at the same time or after a ceasefire or peace agreement. The factors that influence enforcement missions are likely to be different from those that influence more traditional peacekeeping missions where the role of the mission has been agreed upon by the combatants. The following analysis will focus on peacekeeping missions that follow a peace agreement or ceasefire.³

In practice, the agreement to send peacekeepers to a conflict usually occurs simultaneously with the decision by combatants to sign a peace agreement. In most cases, the role of the peacekeepers is explicitly written into the peace agreement and then signed by combatants. For example, in Guatemala, the peace agreement declared, “The ceasefire shall enter into force as of 0000 hours on D-day, the date on which the United Nations verification mechanism shall be in place with full operational capacity”.

Similarly, in one of the UN’s largest and most comprehensive missions in Cambodia, the agreement declared, “UNTAC will exercise the powers necessary to ensure the implementation of this Agreement, including those relating to the organization and conduct of free and fair elections and the relevant aspects of the administration of Cambodia”.

The Comprehensive Peace Agreement signed in Sudan in 2005 stated that “The parties agree to request the United Nations to constitute a lean, effective, sustainable and affordable UN Peace Support Mission to monitor and verify this Agreement and to support the implementation of the Comprehensive Peace Agreement as provided for under Chapter VI of the UN Charter” (p. 106).⁴

In El Salvador, the Chapultepec Agreement 1/16/1992 stated that “The United Nations shall verify compliance with this Agreement … The Government of El Salvador has submitted to the Secretary-General of the United Nations the timetable for implementing the reduction plan referred to in Section 4 of chapter 1 of this agreement. The Secretary-General has made the timetable known to FMLN. The United Nations shall verify compliance with that timetable” (pp. 66–67).⁵

What these examples suggest is that combatants’ decision to sign a peace agreement is contingent upon the UN’s willingness to send peacekeepers. Furthermore, in this pre-peacekeeping period combatants have the ability, both by their own actions and by the actions of allied states, to influence the likelihood of Security Council meetings about the issue. These strategic actions predate the signing of a peace agreement and the deployment of peacekeeping missions, and demonstrate the interconnected nature of the decision of the combatants to sign an agreement with the UN’s willingness to send peacekeepers.

That is, the combatants in Guatemala, Cambodia, South Sudan and El Salvador may not have signed their respective peace agreements had the UN not agreed to undertake the roles described in the peace agreements. Importantly, it is also likely that the UN would not have been willing to send in troops to Guatemala, Cambodia, South Sudan or El Salvador had the combatants not been willing to agree to a comprehensive peace plan, judging by the fact that the security council waited until after the agreements were signed to authorize the respective peacekeeping missions. In other words, it was only after the combatants agreed to accept UN peacekeeping that the UN agreed to send peacekeepers.

Analysts, then, must account for the simultaneous decisions of the combatants agreement to sign a peace treaty and the UN’s agreement to send in peacekeepers. If, indeed, the two are determined simultaneously, how does one isolate the effect that one has on the other? Below, I utilize an instrumental variable, the time after the Black Hawk Down incident, to identify the effect that UN peacekeeping has on the probability that combatants sign a peace agreement.

Data

This study utilizes a dataset based on the Uppsala Conflict Data Program/Peace Research Institute Oslo (UCDP/PRIO) Armed Conflict Dataset Version 4-2011 (Gleditsch et al., 2002). An armed conflict is defined as “a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths”. The unit of analysis is conflict-year. In order to enter the dataset, a conflict must begin in the post-1945 era and either be resolved between 1 January 1989 and 31 December 2008, or still be ongoing at the point of censoring. Only years in which there were 25 battle-related deaths enter the Armed Conflict Dataset.

Data on peace agreements comes from the UN Peacemaker website.⁶ A peace agreement is defined as an agreement signed by two primary opposing parties that concerns the incompatibility driving the conflict, that is, solving, regulating or outlining the resolution process. The UN Peacemaker website contains the full text of over 750 peace agreements and related documents, often in several languages. A dummy variable indicates the years in the conflict in which a peace agreement was signed between the combatants.⁷ For countries that have more than one rebel group operating at the same time, I examined the text of the agreement to determine which groups were signatories. The study draws data on peacekeeping from the UN Department of Peacekeeping’s website. I code the dummy variable as a “1” for every year in which the UN had an ongoing operation in a country.⁸

In order to examine the relationship between peacekeeping missions and peace agreements, I construct the dataset in the following manner. The peace agreement variable is coded as a “1” for each year the combatants signed a peace agreement. Similarly, I code the peacekeeping mission variable as a “1” the year the mission deploys to a conflict.

In order to account for the simultaneous nature of the peace agreement and peacekeeping equations, for each year there is both a peace agreement and a peacekeeping mission deployment, I examine the text of the peace agreement to ensure that the role of the UN peacekeepers was indeed specified in the peace agreement. It could be the case that the combatants sign a peace agreement with no offer of support from the UN, and the UN subsequently sends in peacekeepers to help. However, I have found no instances of this occurring.

For reasons discussed above, the analysis should not include years in which an enforcement mission is present in a country. Therefore, all observations where an enforcement mission is present in the conflict are not included in the analysis. Additionally, observations where the combatants signed a peace agreement, peacekeepers arrived and the combatants continued to fight in subsequent years with the UN peacekeeping mission present are dropped because these cases are no longer “at risk” for a UN peacekeeping mission in the same sense that all other observations in the dataset are. The question for these cases is “Why does the UN continue to deploy troops in the countries?” as opposed to the question of interest in this study, which focuses on the reasons for UN deployment. Therefore, an observation enters the dataset when: (a) the combatants are actively fighting in a given year; and (b) the UN does not currently have a mission deployed in a county to address the conflict with the rebel group.

To illustrate the coding protocol, take the example of the conflict between the government of Angola and the UNITA rebels. In 1991 the combatants signed the Bicesse Accords and the UN deployed the UNAVEM II mission.⁹ Thus, 1991 is coded as having both a peace agreement and peacekeeping mission.

Take another example, of Burundi, where both the CNDD-FDD and Palipehutu-FNL were fighting the government. In November 2003, the CNDD-FDD signed a ceasefire and joined the transitional government. The transition took place without the presence of UN peacekeepers. This can be seen as an example of a peace agreement without a peacekeeping mission. While the CNDD-FDD was joining the government, the “Security Council urged Palipehutu-FNL (Rwasa), the only armed rebel group which had not yet joined the Arusha Agreement, to do so” (United Nations, 2008: 91). Palipehutu-FNL did not join the government, and “in May 2004, acting under the enforcement provisions of the UN Charter, the Security Council authorized the deployment, on 1 June, of the United Nations Operation in Burundi (ONUB)”. This case is an example of an enforcement mission, as the UN sent in peacekeepers even though Palipehutu-FNL was fighting the government at the time.

How does the dataset employed in this study differ from previous analyses, discussed above? First, a major difference is the time frame under study. A civil war enters the dataset in this analysis if the combatants engaged in battle in any year from 1989 to 2008, while Gilligan and Stedman (2003) examine wars active from 1989 to 1997, Fortna (2004) examines wars from 1944 to 1997, Mullenbach (2005) examines wars from 1945 to 2002, Greig and Diehl (2005) examine wars from 1945 to 1999, Walter (2002) examines wars from 1940 to 1992, and DeRouen and Sobek (2004) examine wars from 1944 to 1997.

In terms of structure, Gilligan and Stedman (2003), Fortna (2004) and DeRouen and Sobek (2004) use data that are adapted from Doyle and Sambanis (2000). Mullenbach (2005) uses a similar setup, but includes all conflicts that lasted at least 10 days. The unit of analysis in these four studies is the occurrence of a ceasefire, peace agreement or military victory. The authors then employ duration models or logit/probit models that account for duration dependence, which, as Beck et al. (1998) have shown, are equivalent methods of modeling grouped duration data.

Most similar to the data used in this analysis, Greig and Diehl (2005) use the Regan (2002) civil conflict data. The Regan data have a conflict-month unit of analysis, and Greig and Diehl add peace agreement and peacekeeping variables. Using conflict-month or conflict-year as the unit of analysis, instead of the occurrence of a ceasefire, peace agreement or military victory as in Gilligan and Stedman (2003), Fortna (2004), DeRouen and Sobek (2004) and Mullenbach (2005), allows the analyst to include time-varying covariates. The data and analysis employed in this study, therefore, are similar to those employed in the previous analyses, with the exceptions that the data in this study cover peacekeeping missions in the 2000s, that this analysis includes time-varying covariates, and that this analysis accounts for the endogenous nature of peacekeeping missions.¹⁰

The dataset, therefore, contains peace agreement and peacekeeping deployment variables for each year of the conflict. If the decision to send peacekeepers to a conflict was exogenous, one could simply regress the peacekeeping variable on the peace agreement variable and obtain the effect that peacekeeping missions have on the probability of a peace agreement. However, as argued earlier, peacekeeping missions are not exogenous, and thus a proper analysis requires an instrumental variables model.

Control variables

Ray (2003) argues that scholars of conflict should avoid the over-use of control variables. Following this direction, I present an initial model with no control variables.

Model 2 contains the following control variables: the log of per-capita GDP (lagged), the log of population size, the log of the cumulative number of battle deaths in the conflict and the log of the number of battle deaths from the previous year, the number of veto players in the conflict, and the strength of the rebel group. Income levels and population size have been shown to be significant predictors of civil war duration (Cunningham et al., 2009; DeRouen and Sobek, 2004). In addition, deaths have been linked to UN peacekeeping, with Gilligan and Stedman (2003: 44) calling the relationship between UN peacekeeping and battle deaths “the most robust result in our analysis”. The addition in this analysis is to include the previous year's battle deaths along with the cumulative battle deaths, which turn out to have different substantive effects. Cunningham (2006) has shown that conflicts with multiple “veto players” last longer than conflicts with two “veto players”. The analysis includes a variable denoting the number of ongoing conflicts in the country, to control for the possibility that the conflict dynamics are different in these cases. Cunningham et al. (2009) show that conflicts involving stronger rebel groups are shorter than conflicts with weak rebel groups, finding that rebel strength increases the probability of a peace agreement, among other decisive outcomes. The second model also includes an ordinal measure of rebel strength from the Non-State Actor dataset.¹¹

Additionally, duration dependence needs to be accounted for (Beck et al., 1998). The analysis follows the suggestion of Carter and Signorino (2010) by including variables for time, measured as the number of days since the conflict began, along with its square and cube.

Statistical model

Greene (2008: 814) presents a maximum likelihood instrumental variable estimator for probit models. The model, however, as implemented in STATA’s “ivprobit” command, is not appropriate when an endogenous variable is binary, and thus not appropriate for the binary peacekeeping variable. Following Greene (2008), when the endogenous variable in a probit model is binary, one may utilize a recursive bivariate probit model. The difference between “ivprobit” and the recursive bivariate probit model is essentially a difference in how the equation estimates the endogenous variable. The “ivprobit” procedure amounts to estimating ordinary least squares on the peacekeeping variable and using the residuals from this regression in the peace agreement probit model. On the other hand, the recursive bivariate probit model estimates a probit model on the peacekeeping variable, and then uses the predicted probabilities in the peace agreement probit model.¹² The model is set up as follows (notation from Greene, 2008: 817–826):¹³

y_{1}^{*} = X' β_{1} + y_{2} γ + ϵ_{1}, y_{1} = 1 if y_{1}^{*} > 0, 0 otherwise

y_{2}^{*} = X' β_{2} + instrument δ + ϵ_{2}, y_{2} = 1 if y_{2}^{*} > 0, 0 otherwise

E [ϵ_{1} | x_{1}, x_{2}] = E [ϵ_{2} | x_{1}, x_{2}] = 0

Var [ϵ_{1} | X] = Var [ϵ_{2} | X] = 1

Cov [ϵ_{1}, ϵ_{2} | x_{1}, x_{2}] = ρ

Using the notation above, $y_{1}$ is whether or not combatants sign a peace agreement, whereas $y_{2}$ is whether or not the UN sends a peacekeeping mission to the conflict.

There are two properties that any instrument must satisfy (Greene, 2008: 316): (a) exogeneity—the instrument must be uncorrelated with the disturbances; and (b) relevance—the instrument must correlate with the independent variables (namely, the endogenous variables in X). If a valid instrument can be found that meets these assumptions, one can consistently estimate the parameters in the model.

Identification

In order for an instrument to be valid, it must affect the UN’s willingness to send troops to a conflict, but have no effect on the combatant’s willingness to sign an agreement. The effect of Black Hawk Down was to bring US and UN involvement to a standstill in the mid-1990s. Lawson (2007) argues that UN disengagement:

began with the withdrawal from Somalia in early 1994. This was a period of disengagement driven by the “Somalia Syndrome”. Reeling from the debacle in Somalia, and with the Rwandan genocide already unfolding, Clinton issued Presidential Decision Directive 25 (PDD 25), which sought to strictly limit future UN missions, and especially US participation in them. It listed seven factors that American officials would consider before approving UN operations to be carried out by non-Americans, and six additional factors to be considered if US forces were to participate … The first two considerations in approving even UN operations that excluded US participation were whether the operation would advance US interests, and whether there was a clear threat to international peace and security. (p. 3)

Adebajo (2012) argues that PDD 25 had immediate effects on UN peacekeeping efforts:

Following the Somalia debacle, the administration of US President Bill Clinton placed severe restrictions on the approval of future UN missions through the heavy-handed Presidential Directive 25. Boutros-Ghali’s requests for new UN peacekeeping missions in Burundi and Liberia met with an eloquent silence … Undoubtedly the most tragic consequence of Somalia, however, occurred six months after the killing of the eighteen American soldiers. Washington led the opposition to a UN response to the genocide in Rwanda in which an estimated eight hundred thousand people were massacred in three months in a situation that was tragically and erroneously viewed through a tainted Somali prism. (p. 175)

The effect of PDD 25 on peacekeeping efforts was easily identifiable to observers of UN policy in the late 1990s. An Associated Press article entitled “U.S. peacekeeping policy limiting choices by U.N.” in the Observer-Reporter on 17 October 1997 argues that:

Right or wrong, US policy was the main reason the United Nations did not send a peacekeeping force into the Republic of Congo in time to stop the fighting that forced out the country’s elected president … The same considerations that kept Washington from endorsing UN troops for the Republic of Congo are likely to emerge elsewhere. But without Washington’s participation—either with troops, political support or money—the international security machinery established to keep peace since 1945 doesn’t work. Nobody has come up with an effective alternative, especially in developing countries. The world’s ability to respond quickly to emerging crises broke down in 1993 when 18 US soldiers were killed in Mogadishu, Somalia, and Congress lost interest in offering American troops, money and resources to put out fires in countries most voters don’t even know. Those soldiers who died in the Mogadishu firefight with Somalia militiamen were under US command. But congressional critics blamed their deaths on UN incompetence. Under pressure from Congress, President Clinton announced guidelines for US participation in UN peacekeeping operations, including giving Congress 15 days notification before approving a UN mission. Before giving its approval in the Security Council, where it has a veto, the White House agreed to consider whether the operation would serve US interests. No one envisioned US troops joining the proposed UN force for the Republic of Congo. Nevertheless, US diplomats cited the Clinton guidelines in explaining why they could not support the operation.

In addition, Traub (2006) notes how Black Hawk Down resulted in a shock that led to a lack of UN peacekeeping missions:

In Somalia, the UN—and the Clinton administration as well—discovered how utterly unprepared it was for the new world into which it had so boldly and blithely plunged. Every decision was precipitate, every commitment either grossly inadequate or unsustainable. The UN did not know how to operate in a country with no functioning state, nor how to foster the institutions that might ultimately coalesce into a state. (Neither, of course, did anyone else.) Somalia was a profound shock to the system; the giddy sense of possibility that began with the Gulf War ended with sickening images of the corpses of American soldiers dragged through the dusty streets of Mogadishu. (pp. 38–39, emphasis added.)

In sum, I argue that the Black Hawk Down incident is an exogenous event that: (a) directly influenced UN peacekeeping deployment; and (b) should have no effect on the probability that combatants in other conflicts sign peace agreements other than through its effect on UN peacekeeping. In order to capture this exogenous shock to the system, I create a variable that takes a value of zero prior to Black Hawk Down, and then spikes for the remainder of 1993 and 1994. A loss function then brings the effect of the shock back to zero over time.¹⁴ This variable is a strong predictor of UN peacekeeping missions. Sovey and Green (2011: 199) claim that “although the precise criteria by which to evaluate the weakness of an instrument are subject to debate, the usual rule of thumb is that a single instrumental variable should have an F-statistic of at least 10 in order to avoid appreciable weak instruments bias. In the case of a single instrumental variable, this criterion means that the first-stage t-ratio must be greater than 3.16”. While Sovey and Green provide a rule of thumb for the 2SLS estimator, one might suppose a similar rule of thumb applies to the recursive bivariate probit model employed in this analysis. The z-score for the Black Hawk Down variable in Table 3 is 3.23, even larger than the 3.16 suggested for the 2SLS model, so the instrument should satisfy the “relevance” criteria to be a valid instrument.

Results

Table 2 displays the estimates of two probit models where the signing of a peace agreement is the dependent variable and whether or not the UN has agreed to send peacekeepers is an independent variable, along with the control variables discussed above. These models, like Walter (2002), DeRouen and Sobek (2004) and Greig and Diehl (2005), assume that peacekeeping missions exogenously cause peace agreements. The peacekeeping mission coefficient is positive and statistically significant. This suggests that the UN agreeing to send peacekeepers to a conflict increases the probability that combatants will sign an agreement. If, however, the assumption of exogeneity does not hold, as argued earlier, then the model does not consistently estimate the effect of peacekeeping missions.

Table 2.

Naive probit model—determinants of peace agreements

Variables	(1) peace_agreement	(2) peace_agreement
PK Mission	2.383***	2.320***
PK Mission	(0.343)	(0.354)
Log of GDP Per Capita		−0.029
Log of GDP Per Capita		(0.073)
Log of Population		−0.238**
Log of Population		(0.095)
Cumulative Deaths		0.221
Cumulative Deaths		(0.162)
Previous Year Deaths		−0.592**
Previous Year Deaths		(0.238)
Rebel Strength		0.053
Rebel Strength		(0.117)
Number of Conflicts		0.052
Number of Conflicts		(0.078)
Duration	−0.000**	−0.000
Duration	(0.000)	(0.000)
Duration Squared	0.000**	0.000*
Duration Squared	(0.000)	(0.000)
Duration Cubed	−0.000**	−0.000*
Duration Cubed	(0.000)	(0.000)
Constant	−1.508***	2.822
Constant	(0.161)	(1.799)
Observations	921	790

Robust standard errors in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Tables 3 and 4 display the results of the recursive bivariate probit model. The peacekeeping equation, on the right-hand side of Tables 3 and 4, displays the effects the covariates have on the probability that a peacekeeping mission intervenes in a conflict. As expected, the Black Hawk Down variable is negative and statistically significant. What this means is that the UN was significantly less likely to send UN peacekeeping missions to conflicts in the aftermath of Black Hawk Down. As discussed earlier, the associated levels of significance suggest that the Black Hawk Down variable is a “strong” instrument for peacekeeping.

Table 3.

Recursive bivariate probit model—accounting for endogeneity

Variables	Peace agreement equation	Peacekeeping equation
Black Hawk Down		−0.002**
		(0.001)
PK Mission	−0.269
PK Mission	(0.209)
Constant	−1.570***	−1.705***
Constant	(0.085)	(0.113)
Athrho	1.915***
Athrho	(0.496)
Observations	921

Robust standard errors in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Table 4.

Recursive bivariate probit model—accounting for endogeneity

Variables	Peace agreement equation	Peacekeeping equation
Black Hawk Down		−0.003***
		(0.001)
PK Mission	0.284
PK Mission	(0.611)
Log of Population	−0.218***	−0.071
Log of Population	(0.085)	(0.086)
Log of GDP Per Capita	−0.118	−0.365***
Log of GDP Per Capita	(0.075)	(0.154)
Cumulative Deaths	0.280***	0.308
Cumulative Deaths	(0.128)	(0.218)
Previous Year Deaths	−0.585***	−0.390
Previous Year Deaths	(0.261)	(0.283)
Rebel Strength	0.125	0.104
Rebel Strength	(0.112)	(0.134)
Number of Conflicts	0.028	−0.093
Number of Conflicts	(0.073)	(0.059)
Duration	−0.000	−0.000
Duration	(0.000)	(0.000)
Duration Squared	0.000	0.000
Duration Squared	(0.000)	(0.000)
Duration Cubed	−0.000*	−0.000
Duration Cubed	(0.000)	(0.000)
Constant	3.145***	2.196*
Constant	(1.459)	(1.229)
Athrho	1.176***
Athrho	(0.517)
Observations	790

Robust standard errors in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

The peace agreement equation, on the left-hand side of Tables 3 and 4, displays the effects of the covariates on the probability that the combatants sign a peace agreement. The difference between the results in the peace agreement equations on the left-hand sides of Tables 3 and 4, and the probit models shown in Table 2, is that the latter accounts for the endogenous nature of the peacekeeping variable. After accounting for the endogenous nature of peacekeeping missions, the effect of the peacekeeping variable diminishes significantly. In Table 2 the coefficients of the peacekeeping variables are 2.383 and 2.320, while in Tables 3 and 4 the coefficients are −0.269 and 0.284, respectively. A reduction in size from 2.320 to 0.284 represents an 88% reduction in the size of the peacekeeping coefficient when the model accounts for the endogenous nature of peacekeeping. In addition to a reduction in coefficient size, there is also an increase in the size of the standard error of the peacekeeping variables. This indicates that the peacekeeping variables are no longer close to conventional level of statistical significance. Thus, it appears that the UN’s decision to send peacekeepers to a conflict does not increase the probability of a peace agreement.

Why is it that peacekeeping missions are not related to the probability of peace agreements once we account for endogeneity? Figure 1 provides an answer. While the UN was significantly less likely to send peacekeepers to conflicts in the aftermath of Black Hawk Down in 1993, this did not discourage combatants from signing peace agreements. For example, the Bangui-2 Agreement, signed between the Government of Chad and the Committee of National Revival for Peace and Democracy on 10 August 1994, was not supported by a UN peacekeeping mission. Similarly, the Front for the Restoration of Unity and Democracy in Djbouti signed the Accord de Paix et de la Reconciliation Nationale on 26 December 1994, also without the support of a UN peacekeeping mission. In fact the number of peace agreements in conflicts appears to be relatively stable across time, despite the fact that the UN’s willingness to send peacekeepers varies significantly with time.¹⁵

Figure 1.

Number of peace agreements, by year.

Regarding the covariates, several interesting results emerge. In Table 3, the log of population and log of GDP variables are negative and statistically significant in both the first and second stage equations. What this means is that the UN is less likely to send peacekeepers to large, rich countries; and large, rich countries are less likely to settle their conflict. Just as Gilligan and Stedman (2003) find, peacekeepers are more likely to go to conflicts with higher levels of cumulative battle deaths. Interestingly, countries are more likely to sign peace agreements with higher levels of cumulative battle deaths, but are less likely to sign agreements if those battle deaths occurred in the previous year. Thus, it looks like peace agreements are most likely to occur in countries that experienced high levels of violence at one time, but then went through a period of decline in violence.

Turning to Table 4, it appears that the number of ongoing conflicts in a country and the strength of the rebel groups have no significant effect on UN peacekeeping or the likelihood of a peace agreement. The lack of substantive findings is probably due to the high levels of collinearity between the rebel strength, number of conflicts, and size of the population variables. In fact, if the model in Table 4 does not contain the rebel strength variable, then increasing the number of ongoing conflicts has a negative effect on peacekeeping deployment (p = 0.076). Furthermore, if the model in Table 4 does not contain the log of population variable, the number of ongoing conflicts has a negative and statistically significant effect (p = 0.003) on the peacekeeping deployment, and the rebel strength has a positive and statistically significant effect on the probability of a peace agreement (p = 0.026), which is consistent with the findings in Cunningham et al. (2009).

Discussion

If the UN’s promise to send peacekeepers to a conflict does not increase the probability that combatants sign a peace agreement, as argued here, what implications does this finding have for the study of civil wars? First, as discussed earlier, the finding suggests that comparing conflicts that did receive UN peacekeeping missions with those that did not is possible, after controlling for other covariates.

Second, the finding speaks to the debate about the causes of conflict (Fearon, 1995; Powell, 2004), and the manner in which civil conflicts are resolved (Fearon, 2004; Filson and Werner, 2002; Slantchev, 2003; Smith and Stam, 2004; Zartman, 1985). Recent research argues that peacekeeping missions are effective at maintaining peace in the aftermath of civil conflict (Doyle and Sambanis, 2000; Fortna, 2008; Gilligan and Sergenti, 2008; Walter, 2002). This finding, coupled with the finding from this research that peacekeers do not increase the probability of peace agreements, suggests that some of the potential causal mechanisms for the effectiveness of peacekeeping missions may be more plausible than others.

Fortna (2008: 86) argues that “peacekeepers can have a causal, rather than spurious, effect on the stability of peace if (1) they reduce the likelihood of aggression by raising the costs of war or the benefits of peace for the peacekept; (2) they disrupt spirals of fear and security dilemmas by reducing belligerents’ uncertainty about each other’s actions and intentions; (3) they prevent accidents from occurring or control them so that they do not escalate to war; or (4) they can deter or prevent one side from reneging on a political deal and excluding the other from power”.

If it is true that peacekeeping is effective, but that it does not increase the probability of a peace agreement, then it is probable that peacekeeping missions do not affect the combatants’ cost–benefit calculations when deciding whether or not to settle a conflict. If peacekeepers alter the cost–benefit calculations of the parties, then this effect should be reflected in the analysis. For example, if a rebel group is worried about being excluded from power in the future, and peacekeepers lower the probability of exclusion occurring, then the rebel group should be more likely to sign a peace agreement when peacekeepers are present. The finding here, however, is that the presence of peacekeepers does not alter the willingness of combatants to sign peace agreements. Fortna’s points 1 and 4 suggest that peacekeepers alter the cost–benefit calculus of the parties by preventing one of the parties from restarting the war or excluding the other party from power when it becomes beneficial to them. The finding from this analysis casts doubt upon arguments that point to these as reasons why peacekeepers are effective.

The findings of this study suggest that points 2 and 3 may do a better job of explaining the reasons peacekeeping missions are successful because these explanations do not require the peacekeepers to alter the cost–benefit calculus of the combatants. These points argue that peacekeepers can provide valuable information about the intentions of the other party, meaning that fears do not spiral out of control, and accidents do not derail the peace process. While the findings in this research do not directly test these four causal mechanisms, the logical implications of the findings suggest that peacekeepers may be most effective for their information-gathering and accident-prevention roles.

Finally, while this study focuses on Black Hawk Down’s effect on the probability of peace agreements, the incident also had a large effect on the strategic actions of combatants once peacekeepers are on the ground. Combatants, knowing the reluctance of contributing countries to suffer casualties, may attack peacekeepers in an effort to remove them. This was, seemingly, the reason Hutu extremists attacked Belgian peacekeepers in Rwanda, and the strategy worked. “Just as Belgian political leaders had in the past used public opinion as a reason for seeking broader involvement in Rwanda, so now they relied on it to try to justify their withdrawal. They referred to the ‘great emotion’ caused in Belgium by the loss of the peacekeepers and to a public opinion ‘traumatized’ by their deaths” (Desforges, 1999: 474). Indeed, because of the Black Hawk Down incident, it is now commonly assumed that countries will not become involved in humanitarian crises for fear that an operation will cross the “Mogadishu Line”, and will turn into combat operations.¹⁶

Footnotes

Acknowledgements

I would like to thank Alastair Smith, Michael Gilligan, Bruce Bueno de Mesquita, Neal Beck, Shanker Satyanath, Bernd Beber, Robin Harding, Andrew Little and MacKenzie Canniff for their helpful comments and suggestions. Data and replication material can be found at:

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Notes

References

Adebajo

(2012) UN Peacekeeping in Africa. Boulder, CO: Lynne Rienner.

Angrist

Krueger

(2001) Instrumental variables and the search for identification. Journal of Economic Perspectives 15: 69–85.

Ashbrook

(1995) UN effects everywhere turn to dust—Downed helicopter in Somalia doomed a “new world order”. Boston Globe, 30 April.

Balch-Lindsay

Enterline

(2000) Killing time: The world politics of civil war duration, 1820–1992. International Studies Quarterly 44(4): 615–642.

Balch-Lindsay

Enterline

Joyce

(2008) Third-party intervention and the civil war process. Journal of Peace Research 45(3): 345.

Beck

Katz

Tucker

(1998) Beyond ordinary logit: Taking time seriously in binary time-series-cross-section models. American Journal of Political Science 42(4): 1260–1288.

Carter

Signorino

(2010) Back to the future: Modeling time dependence in binary data. Political Analysis 18(3): 271–292.

Cunningham

(2006) Veto players and civil war duration. American Journal of Political Science 50(4): 875–892.

Cunningham

Gleditsch

Salehyan

(2009) It takes two. Journal of Conflict Resolution 53(4): 570.

10.

DeRouen

Sobek

(2004) The dynamics of civil war duration and outcome. Journal of Peace Research 41(3): 303–320.

11.

Desforges

(1999) Leave none to tell the story: Genocide in Rwanda. New York: Human Rights Watch.

12.

Doyle

Sambanis

(2000) International peacebuilding: A theoretical and quantitative analysis. The American Political Science Review 94(4): 779–801.

13.

Fearon

(1995) Rationalist explanations for war. International Organization 49(3): 379–414.

14.

Fearon

(2004) Why do some civil wars last so much longer than others? Journal of Peace Research 41(3): 275–301.

15.

Filson

Werner

(2002) A bargaining model of war and peace: Anticipating the onset, duration, and outcome of war. American Journal of Political Science 46(4): 819–837.

16.

Fortna

(2004) Does peacekeeping keep peace? International intervention and the duration of peace after civil war. International Studies Quarterly 48: 269–292.

17.

Fortna

(2008) Does Peacekeeping Work?: Shaping Belligerents’ Choices after Civil War. Princeton, NJ: Princeton University Press.

18.

Gilligan

Sergenti

(2008) Do UN interventions cause peace? Using matching to improve causal inference. Quarterly Journal of Political Science 3(2): 89–122.

19.

Gilligan

Stedman

(2003) Where do the peacekeepers go? International Studies Review 5(4): 37–54.

20.

Gleditsch

Wallensteen

Eriksson

Sollenberg

Strand

(2002) Armed conflict 1946–2001: A new dataset. Journal of Peace Research 39(5): 615–637.

21.

Greene

(2008) Econometric Analysis. Englewood Cliffs, NJ: Pearson/Prentice Hall.

22.

Greig

Diehl

(2005) The peacekeeping–peacemaking dilemma. International Studies Quarterly 49(4): 621–645.

23.

King

Zeng

(2007) When can history be our guide? The pitfalls of counterfactual inference. International Studies Quarterly 51(1): 183–210.

24.

Lawson

(2007) US Africa policy since the Cold War. Strategic Insights VI(1).

25.

Mullenbach

(2005) Deciding to keep peace: An analysis of international influences on the establishment of third-party peacekeeping missions. International Studies Quarterly 49(3): 529–556.

26.

Powell

(2004) The inefficient use of power: Costly conflict with complete information. American Political Science Review 98(2): 231–241.

27.

Ray

(2003) Explaining interstate conflict and war: What should be controlled for? Conflict Management and Peace Science 20(2): 1–31.

28.

Regan

(2002) Third-party interventions and the duration of intrastate conflicts. Journal of Conflict Resolution 46(1): 55.

29.

Sambanis

Doyle

(2007) No easy choices: Estimating the effects of United Nations peacekeeping (response to King and Zeng). International Studies Quarterly 51(1): 217–226.

30.

Slantchev

(2003) The principle of convergence in wartime negotiations. American Political Science Review 97(4): 621–632.

31.

Smith

Stam

(2004) Bargaining and the nature of war. Journal of Conflict Resolution 48(6): 783.

32.

Sovey

Green

(2011) Instrumental variables estimation in political science: A readers’ guide. American Journal of Political Science 55(1): 188–200.

33.

Traub

(2006) The Best Intentions: Kofi Annan and the UN in the Era of American World Power. New York: Farrar Straus & Giroux.

34.

United Nations (2008) United Nations Today. New York: United Nations Department of Public Information.

35.

Walter

(2002) Committing to Peace. Princeton, NJ: Princeton University Press.

36.

Wooldridge

(2002) Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press.

37.

Zartman

(1985) Ripe for Resolution: Conflict and Intervention in Africa. Oxford: Oxford University Press.